sql server data mining:
Hello all, My Name is Aric Katz and i am working AT Rafael (Israel), that is an aero-space and defense company, as a head developer on project management systems parallel to my job I am working on my PhD and I am trying to implement some mining models to figure out the trends of arrivals to a hospitals X-Ray department, I separated the data to: 1. year 2. quarter 3. month 4. weak 5. day 6. hour 7. the primary data = number of patients I am trying to find the trends by time periods of different amounts of arrivals, any recommendation about what kind of alg' to use and how will be much appreciated.
You can use time series for forecasting, but sometimes you may also get good results to convert a forecasting problem to regression or classification (discretized). For example, using a decision tree algo to predict number of patients based on the attributes Year, Quater, Month, Week, Day, Hour. ZhaoHui [quoted text, click to view] <aricka@rafael.co.il> wrote in message
news:f1bcaf30-9a13-46f8-85af-d81d10a3204b@msnews.microsoft.com...
> Hello all,
> My Name is Aric Katz and i am working AT Rafael (Israel), that is an
> aero-space and defense company, as a head developer on project management
> systems
> parallel to my job I am working on my PhD and I am trying to implement
> some mining models to figure out
> the trends of arrivals to a hospitals X-Ray department, I separated the
> data to:
> 1. year
> 2. quarter
> 3. month
> 4. weak
> 5. day
> 6. hour
> 7. the primary data = number of patients
>
> I am trying to find the trends by time periods of different amounts of
> arrivals,
> any recommendation about what kind of alg' to use and how will be much
> appreciated.
>
> thanks in advance, Aric Katz.
I would use a regression tree model to predict # patients, but make sure you use relative inputs instead of absolute inputs. e.g. Quarter of Year Month of Year Week of Month Day of Week Hour of Day You may also want to add Month of Quarter Week of Year Day of Month etc. Also a field like "Holiday" may be good as well. If you can assume that patterns are independent of year, you should also normalize by the year. That is, if you had data for 2003-2005, and 2003 and 910 total visits, 2004 had 1000, and 2005 had 1100, you would multiply all 2003 counts by 1.1 and all 2005 counts by 0.9 so that the overall counts would be the same for each year. The resultant model will give you a piecewise linear regression describing patterns for the visits [quoted text, click to view] >
[quoted text, click to view] > Hello all,
> My Name is Aric Katz and i am working AT Rafael (Israel), that is an aero-space and defense company, as a head developer on project management systems
> parallel to my job I am working on my PhD and I am trying to implement some mining models to figure out
> the trends of arrivals to a hospitals X-Ray department, I separated the data to:
> 1. year
> 2. quarter
> 3. month
> 4. weak
> 5. day
> 6. hour
> 7. the primary data = number of patients
>
> I am trying to find the trends by time periods of different amounts of arrivals,
> any recommendation about what kind of alg' to use and how will be much appreciated.
>
I would use a regression tree model to predict # patients, but make sure you use relative inputs instead of absolute inputs. e.g. Quarter of Year Month of Year Week of Month Day of Week Hour of Day You may also want to add Month of Quarter Week of Year Day of Month etc. Also a field like "Holiday" may be good as well. If you can assume that patterns are independent of year, you should also normalize by the year. That is, if you had data for 2003-2005, and 2003 and 910 total visits, 2004 had 1000, and 2005 had 1100, you would multiply all 2003 counts by 1.1 and all 2005 counts by 0.9 so that the overall counts would be the same for each year. The resultant model will give you a piecewise linear regression describing patterns for the visits [quoted text, click to view] >
[quoted text, click to view] > Hello all,
> My Name is Aric Katz and i am working AT Rafael (Israel), that is an aero-space and defense company, as a head developer on project management systems
> parallel to my job I am working on my PhD and I am trying to implement some mining models to figure out
> the trends of arrivals to a hospitals X-Ray department, I separated the data to:
> 1. year
> 2. quarter
> 3. month
> 4. weak
> 5. day
> 6. hour
> 7. the primary data = number of patients
>
> I am trying to find the trends by time periods of different amounts of arrivals,
> any recommendation about what kind of alg' to use and how will be much appreciated.
>
thanks for all of the replies, i am going to try all of your greate ideas, i did a minor change, i am looking for arrivals so i built a query that will return only the hours where the arrivals were positive so that left me with less but more focused data that differs from the original data by being the delta of patients between two following hours can anyone recomend of a good tutorial or webcast that displays the use of the regression tree model? agian thanks alot for of the responses> [quoted text, click to view] > I would use a regression tree model to predict # patients, but make sure you use relative inputs instead of absolute inputs. e.g.
>
>
> Quarter of Year
> Month of Year
> Week of Month
> Day of Week
> Hour of Day
>
> You may also want to add
>
> Month of Quarter
> Week of Year
> Day of Month
> etc.
> Also a field like "Holiday" may be good as well.
>
> If you can assume that patterns are independent of year, you should also normalize by the year. That is, if you had data for 2003-2005, and 2003 and 910 total visits, 2004 had 1000, and 2005 had 1100, you would multiply all 2003 counts by 1.1 and all 2005 counts by 0.9 so that the overall counts would be the same for each year.
>
> The resultant model will give you a piecewise linear regression describing patterns for the visits
>
> >
>
>
> > Hello all,
> > My Name is Aric Katz and i am working AT Rafael (Israel), that is an aero-space and defense company, as a head developer on project management systems
> > parallel to my job I am working on my PhD and I am trying to implement some mining models to figure out
> > the trends of arrivals to a hospitals X-Ray department, I separated the data to:
> > 1. year
> > 2. quarter
> > 3. month
> > 4. weak
> > 5. day
> > 6. hour
> > 7. the primary data = number of patients
> >
> > I am trying to find the trends by time periods of different amounts of arrivals,
> > any recommendation about what kind of alg' to use and how will be much appreciated.
> >
Don't see what you're looking for? Try a search.
|
